arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2402.10376cs.LGcs.CV
收藏

用稀疏线性概念嵌入(SpLiCE)解释CLIP

Interpreting CLIP with Sparse Linear Concept Embeddings (SpLiCE)

Harvard University(哈佛大学)

Usha Bhalla, Alex Oesterling, Suraj Srinivas, Flavio P. Calmon, Himabindu Lakkaraju

AI总结:

本文提出SpLiCE方法,将CLIP嵌入分解为稀疏线性概念组合,实现任务无关的可解释性,无需训练即可解释甚至替代密集表示,并用于检测虚假相关性和模型编辑。

摘要:

CLIP嵌入已在广泛的多模态应用中展现出卓越性能。然而,这些高维、密集的向量表示不易解释,限制了我们对CLIP丰富结构的理解及其在需要透明度的下游应用中的使用。在这项工作中,我们展示了CLIP潜在空间的语义结构可以被利用来提供可解释性,从而将表示分解为语义概念。我们将此问题表述为稀疏恢复问题,并提出一种新方法——稀疏线性概念嵌入(SpLiCE),用于将CLIP表示转换为人类可解释概念的稀疏线性组合。与以往工作不同,SpLiCE是任务无关的,可以在无需训练的情况下用于解释甚至替代传统的密集CLIP表示,在保持高下游性能的同时显著提高其可解释性。我们还展示了SpLiCE表示的重要用例,包括检测虚假相关性和模型编辑。

英文摘要:

CLIP embeddings have demonstrated remarkable performance across a wide range of multimodal applications. However, these high-dimensional, dense vector representations are not easily interpretable, limiting our understanding of the rich structure of CLIP and its use in downstream applications that require transparency. In this work, we show that the semantic structure of CLIP's latent space can be leveraged to provide interpretability, allowing for the decomposition of representations into semantic concepts. We formulate this problem as one of sparse recovery and propose a novel method, Sparse Linear Concept Embeddings, for transforming CLIP representations into sparse linear combinations of human-interpretable concepts. Distinct from previous work, SpLiCE is task-agnostic and can be used, without training, to explain and even replace traditional dense CLIP representations, maintaining high downstream performance while significantly improving their interpretability. We also demonstrate significant use cases of SpLiCE representations including detecting spurious correlations and model editing.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑